SUNET: Efficiently Scaling a 100-GB Converged Optical IP Network
Mar 22, 2016
This is a guest blog post. Views expressed in this post are original thoughts posted by Per Nihlén, chief technology officer at SUNET. These views are his own and in no way do they represent the views of the company he works for.
At SUNET, we provide high-speed data services to universities and colleges across Sweden, connecting over 30 universities and colleges. We support about half a million users, operating in both research and production environments.
We'd been running our previous network for about ten years, but the optical transport equipment was approaching its end-of-life, and the contract with our dark-fiber supplier was coming up for renewal. We knew traffic would continue to increase, so we needed a more efficient way to grow and contain our costs. We took this opportunity to use a brand-new approach for our network, multilayer optical and IP convergence.
Optical integration into the routing layer makes perfect sense. We no longer need transponders sitting back-to-back with optics in the routers - and we don't need redundant transponders lying around to do re-routing if necessary. Simplifying the network like this saves component costs and improves availability because there are fewer components that can fail. Because the router has an inherent understanding of the fiber conditions, it can re-route around failures before the fiber has even broken.
We used Juniper Networks to supply the technology, using their Juniper Networks® MX 3D Universal Edge Routers. They are connected to over approximately 8,000 km of dark fiber at speeds of 100 GB, using Dense Wavelength Division Multiplexing (DWDM) coherent optical interfaces, the first deployment of their type anywhere in the world. The optical domain is provided using equipment from ADVA Optical Networking, including DWDM equipment, reconfigurable optical add-drop multiplexers (ROADM) and In-Line Amplifiers (ILA). We also avoided using any classical optical DWDM filters in the network and deployed so called colorless ROADM systems, creating a fully gridless domain. When we need to run our network at even higher speeds, such as a terabit per second or more, we might need to use more optical spectrum per channel than today's 50GHz ITU standard. By avoiding the use of classical optical filters, we can do this without touching the optical network at all. But to make this work we need tunable optical interfaces that don't scan across the whole channel range and interfere with other optical paths. With Juniper we've been able to deploy a network that doesn't use any classical optical filters, giving us massive flexibility and the potential to scale.
Integrating IP and optical layers like that also means that we can provision optical connectivity directly from the router in a very flexible way, setting optical launch power and DWDM frequency. And now we'll build in further automation using the NETCONF (Network Configuration Protocol) YANG modeling language for configuring network equipment, including both the routers and the optical layer. For us, automating network configuration like this is a practical example of SDN.
We should never be the bottleneck for any research project. Now we have the capacity we need to support our users delivering research projects, distance learning, cloud based services and huge volumes of data transfer for institutions such as the Swedish National Facility for Radio Astronomy, ESS and Eiscat 3D. Once we've fully deployed the new network, all the universities in Sweden will be connected by two 100-GB capacity uplinks.
And network availability is as critical as performance. The cost of research experiments can often be huge compared to the network costs. The new network is not only far more resilient, it's also more efficient. We've been able to add more routers than we had before, so now we can route traffic more optimally between universities that are close to each other. And as a research network, we have to interoperate with other institutions around the world. So being able to deploy all of these using open standards was also an important result.
If we'd taken a traditional network approach, we would have spent far more money and achieved less. Now we have a network that's at least ten times faster than we used to have, has greater resiliency and actually has lower costs than it had before, all down to the integrated optics. Whatever the future holds, we know our customers will need more bandwidth, and now we can scale to support them and do it in a cost effective way.